Safety Performance Functions for Rural Two-Lane County Road Segments - PowerPoint PPT Presentation

Safety Performance Functions for Rural Two-Lane County Road Segments Steven Y. Stapleton Anthony Ingle Meghna Chakraborty Timothy J. Gates, Ph.D., P.E., PTOE Peter T. Savolainen, Ph.D., P.E. June 22, 2018

Background: Current SPF Limitations � HSM SPFs for rural two-lane segments (2U) based off data from 1,331 sites on state- maintained segments in MN and WA � Predictive accuracy varies from state to state � Differences in geography, design, driver behavior, etc. � HSM recommends recalibration or full re- estimation of SPFs using local data � Neither HSM nor state specific SPFs use data from county-maintained roadways � Rural, low-volume, gravel, non-federal aid 2

Background: Michigan Rural Road Statistics � All 83 counties in Michigan maintain road network � 84,000 miles of rural highway � 69% of Michigan’s total roadway mileage � > 72,000 miles of county owned rural highways � 86% of Michigan’s rural highway mileage � 60% of Michigan’s total roadway mileage � 4th largest county road system in US � 60% of rural crashes occur on county roads 3

Objective � Develop fully-specified SPFs for county-maintained highway segments in Michigan � Rural, 55 mph � Paved and gravel � Federal aid (FA) and non-federal aid (non-FA) � Broad statewide geographic distribution 4

Data Collection: Geographic Scope � 29 counties � All regions of the state � Excluded all incorporated areas and census designated places � Min. segment length: 0.2 mi 5

Data Collection: Data Sources Data Fed Aid Segments Non-Fed Aid Segments AADT MDOT database County road commissions, RPCs Lane Width Manual review (Google Earth) Shoulder Width Manual review (Google Earth) Driveway Counts Manual review (Google Earth) Horizontal Curvature MSU database Crashes Michigan State Police database 6

Summary Statistics AADT Low Volume Non-FA Paved FA Paved Non-FA Paved Gravel Min 251 3 4 3 Max 12,781 12,628 399 399 Mean 1,789 572 133 207 Annual Midblock Segment Crashes (per Mile) Low-Volume Non-FA Paved FA Paved Non-FA Paved Gravel Non-Deer PDO Crashes 0.43 0.15 0.07 0.09 Non-Deer FI Crashes 0.17 0.06 0.03 0.04 Deer Crashes 1.10 0.37 0.17 0.08 Total Crashes 1.70 0.58 0.27 0.21 Deer Crashes, % of Total 64.7% 63.8% 63.2% 38.8% HSM Data from Washington State included 12% animal crashes

Summary Statistics Low Volume Non-FA (Paved and Gravel) Paved FA Paved Non-FA Major Collector: 1% Minor Collector: 6% Minor Minor Arterial: Collector: 12% 19% Major Collector: 88% Local Road: 80% Local Road: 94% Average Segment Average Segment Average Segment Length: Length: Length: 0.52 mi 0.58 mi 0.57 mi Total Length: Total Length: Total Length: 3,616 mi 1,398 mi 1,984 mi 8 8

Analytical Method � Crashes are non-negative integers � Poisson assumption: Variance equals mean � Crash data typically over-dispersed -> Negative binomial � ​𝜇↓𝑗 =exp(β ​𝑌↓𝑗 + ​ ε ↓𝑗 ) • X i = vector of estimable parameters • β = parameter estimate • ε = gamma distributed term with mean 0 and variance α � Multiple counties with different design standards � County-specific random effect (panel data) � ​𝜇↓𝑗𝑘 =exp​(β ​𝑌↓𝑗𝑘 + ​ ε ↓𝑗𝑘 + ​𝜃↓𝑗𝑘 ) • j = county panel indicator • 𝜃 = gamma distributed term with = gamma distributed term with mean 0 and variance α 9

Model Interpretation � Estimation of crashes from RENB takes the form: Where: 𝑂 =exp​( ​𝛾↓𝑝 + ​𝛾↓𝑗 ​𝑌↓𝑗 ) • β o = intercept term, • X i = vector of estimable parameters, • β i = parameter estimate � This can be simplified to the following: 𝑂 = ​𝑓↑​𝛾↓𝑝 ∗ 𝑇𝑓𝑕𝑛𝑓𝑜𝑢 𝑀𝑓𝑜𝑕𝑢ℎ ∗ 𝐵𝐵𝐸​𝑈↑​𝛾↓ 1 ∗ 𝐷𝑁​𝐺↓𝑗 Where: • CMF i = exp( β i ) � Interpretation of CMFs: � CMF>1: Increase in crashes � CMF<1: Decrease in crashes • Percent reduction in crashes: 100∗(1− 𝐷𝑁𝐺 ) • Percent increase in crashes: 100∗( 𝐷𝑁𝐺 −1) 10

SPF Functional Form: Paved Federal Aid Segments ​𝑂↓𝑁𝐽𝐸𝐸𝐹 _ 𝑢𝑝𝑢 = ​𝑓↑ −5.99 ∗ (𝑇𝑓𝑕𝑛𝑓𝑜𝑢 𝑀𝑓𝑜𝑕𝑢ℎ) ∗ 𝐵𝐵𝐸​𝑈↑ 0.71 ∗ 𝐷𝑁𝐺 ​𝑂↓𝑁𝐽𝐸𝐸𝐹 _ 𝐺𝐽 = ​𝑓↑ −7.43 ∗ (𝑇𝑓𝑕𝑛𝑓𝑜𝑢 𝑀𝑓𝑜𝑕𝑢ℎ) ∗ 𝐵𝐵𝐸​𝑈↑ 0.759 ∗ 𝐷𝑁𝐺 Where: • MIDDE_tot = Total (KABCO) midblock non-deer crashes Parameter KABCO CMF FI CMF • MIDDE_FI = Fatal and injury midblock non-deer crashes Presence of curve with design speed < 55 mph 1.56 1.54 Lane width >12 S Not Significant 0.73 10 to 15 driveways per mile 1.07 Not Significant 15 driveways per mile or greater 1.15 Not Significant 11

Results: Paved Federal Aid Segments 3.5 3.5 KABCO Crashes FI Crashes Curve 3 3 EsVmated Mean Non-Deer Midblock Crashes per Mile EsVmated Mean Non-Deer Midblock Crashes per Mile 2.5 2.5 15+ driveways 10-15 driveways Base 2 2 1.5 1.5 Curve 1 1 Base Lane Width >12 ft 0.5 0.5 0 0 250 2250 4250 6250 8250 10250 12250 250 2250 4250 6250 8250 10250 12250 12 AADT AADT

SPF Functional Form: Paved Non Federal Aid Segments ​𝑂↓𝑁𝐽𝐸𝐸𝐹 _ 𝑢𝑝𝑢 = ​𝑓↑ −6.23 ∗ (𝑇𝑓𝑕𝑛𝑓𝑜𝑢 𝑀𝑓𝑜𝑕𝑢ℎ) ∗ 𝐵𝐵𝐸​𝑈↑ 0.73 ∗ 𝐷𝑁𝐺 ​𝑂↓𝑁𝐽𝐸𝐸𝐹 _ 𝐺𝐽 = ​𝑓↑ −7.94 ∗ (𝑇𝑓𝑕𝑛𝑓𝑜𝑢 𝑀𝑓𝑜𝑕𝑢ℎ) ∗ 𝐵𝐵𝐸​𝑈↑ 0.787 ∗ 𝐷𝑁𝐺 Where: • MIDDE_tot = Total (KABCO) midblock non-deer crashes Parameter KABCO CMF FI CMF • MIDDE_FI = Fatal and injury midblock non-deer crashes Presence of curve with design speed < 55 mph 1.45 1.76 13

Functional Form of SPF: Low Volume Non Federal Aid Segments ​𝑂↓𝑁𝐽𝐸𝐸𝐹 _ 𝑢𝑝𝑢 = ​𝑓↑ −5.55 ∗ (𝑇𝑓𝑕𝑛𝑓𝑜𝑢 𝑀𝑓𝑜𝑕𝑢ℎ) ∗ 𝐵𝐵𝐸​ 𝑈↑ 0.674 ∗ 𝐷𝑁𝐺 ​𝑂↓𝑁𝐽𝐸𝐸𝐹 _ 𝐺𝐽 = ​𝑓↑ −6.19 ∗ (𝑇𝑓𝑕𝑛𝑓𝑜𝑢 𝑀𝑓𝑜𝑕𝑢ℎ) ∗ 𝐵𝐵𝐸​𝑈↑ 0.584 ∗ 𝐷𝑁𝐺 Where: • MIDDE_tot = Total (KABCO) midblock non-deer crashes Parameter KABCO CMF FI CMF • MIDDE_FI = Fatal and injury midblock non-deer crashes Presence of curve with design speed < 55 mph 1.96 2.03 Paved surface 0.67 0.58 14

Comparison of Base SPFs 3.5 HSM 3 Mean EsVmated KABCO Non-Deer Crashes per Mile 2.5 FA 2 MDOT Non-FA 1.5 1 0.5 0 250 2250 4250 6250 8250 10250 12250 15 AADT

Curve CMF FA Curve 3 Non-FA Curve Mean EsVmated KABCO Non-Deer Crashes per Mile 2.5 MDOT Curve FA Base 2 MDOT Base Non-FA Base 1.5 1 0.5 0 250 2250 4250 6250 8250 10250 12250 16 AADT

Pavement Surface CMF (Low Volume) 0.25 Gravel 0.2 Mean EsVmated KABCO Non-Deer Crashes per Mile 0.15 Paved HSM 0.1 MDOT 0.05 0 17 0 50 100 150 200 250 300 350 400 AADT

Conclusions � Paved non-FA segments performed the best � FA segments showed similar performance � HSM models more linear than county models • Over-predict at high volumes, under-predict at low volumes � MDOT models under-predict relative to county models � Gravel roads showed the highest crash occurrence rates, particularly for PDO crashes and curved segments � Reduced surface friction � Poorer maintenance � Less aggressive snow removal � Reduced roadside clear zone 18 � Lower speeds?

Conclusions � Presence of a horizontal curve <55 mph was positively correlated with crashes across all segment types � Speeds too fast � Limited sight distance � Driveways increased total crash occurrence on FA � 10-15 7% increase � 15+ 15% increase � Effects of lane width and shoulder width were mostly inconclusive � Lane width significant for FI crashes on federal aid segments (most similar to state highways) 19

Limitations � Dataset limited to Michigan � Gathering data was a laborious process – not feasible for most states to develop their own county SPFs • Calibration is a possibility � Roadside data unavailable � Clear zone, foreslope, etc. � Inconsistent design and maintenance practices between counties � Cross-sectional study � Use pavement type CMF with caution (not B&A) � Crash reductions reflect equal AADTs • If a gravel road is paved, will traffic migrate to this road? 20

Questions? Timothy J. Gates, Ph.D., P.E. Associate Professor Michigan State University Department of Civil and Environmental Engineering 517-353-7224 gatestim@msu.edu � MDOT report available by Googling “MDOT SPR 1645” Funding provided by: 21